Note: This report is based on a close reading of the asyncio code, rather than having experienced defective behaviour in test or production.

The documentation for BaseTransport.close() states: "Close the transport. If the transport has a buffer for outgoing data, buffered data will be flushed asynchronously."

The _UnixWritePipeTransport._close method, which is called by close() contains this code in Python 3.5.1

    def _close(self, exc=None):
        self._closing = True
        if self._buffer:
            self._loop.remove_writer(self._fileno)
        self._buffer.clear()
        self._loop.remove_reader(self._fileno)
        self._loop.call_soon(self._call_connection_lost, exc)

In this context, _buffer is a list of bytes objects comprising yet-to-be-written data. Note that close() removes the writer if _buffer is *not* empty, so the buffered data will never be written.  I believe this conditional should be inverted, so the writer is removed only if the buffer is *empty*. So:

        if not self._buffer:
            self._loop.remove_writer(self._fileno)

Arguably though, a more robust and easier to read test would be to call get_write_buffer_size(), like this:

        if self.get_write_buffer_size() == 0:
            self._loop.remove_writer(self._fileno)

Note that in the event that the writer is not removed by _close(), the writer will
remove itself when the buffer does eventually become empty.

Assuming my analysis is correct, and a fix is made to close(), then abort() will also need to be modified to *not* flush buffered writes, as the documentation for abort() states "Close the transport immediately, without waiting for pending operations to complete. Buffered data will be lost."